Generative AI in Higher Education Teaching & Learning: Principles for Ethical AI Adoption

Generative AI in Higher Education Teaching & Learning Principles for Ethical AI Adoption Contributors James O’Sullivan Colin Lowry Ross Woods Tim Conlon Acknowledgements Alan Smeaton (Government of Ireland AI Advisory Council), Barry O’Sullivan (Government of Ireland AI Advisory Council), Susan Leavy (Government of Ireland AI Advisory Council), Caoimhe Hope (Department of Further and Higher Education, Research, Innovation and Science), Anne RibaultO’Reilly (Department of Further and Higher Education, Research, Innovation and Science), Ana Rocha (Higher Education Authority), Imma Zoppi (Higher Education Authority), Paul O’Donovan (University College Cork), Joseph Feller (University College Cork), Danielle Duignan (AHEAD), Jim O’Mahony (Munster Technological University), Denise Mac Giolla Rí (Technological University of the Shannon), Justin Tonra (University of Galway), Brian Marrinan (Journey Partners), Clyde Hutchinson (Journey Partners), Mary-Claire Kennedy (University of Limerick), Maria Murphy (Maynooth University), Derek Dodd (Technological University Dublin), Pauline Rooney (Trinity College Dublin), Loretta Goff (University College Cork), Roisín Morris-Drennan (Quality and Qualifications Ireland), Liam Fogarty (University College Dublin), Barbara Whelan (Microsoft), Johanna Archbold (Atlantic Technological University), Áine Clarke (Ibec), Marie Clarke (University College Dublin), Bryan O’Mahony (Aontas na Mac Léinn in Éirinn), Nessa McEniff (Learnovate), Rosemary Day (Mary Immaculate College), Leo Casey (National College of Ireland), Alison Cook-Sather (Bryn Mawr College), Jan McArthur (Lancaster University), Paul McSweeney (University College Cork), Áine Ní Shé (Munster Technological University), Frances O’Connell (Technological University of the Shannon), Emma Muldoon Ryan (Aontas na Mac Léinn in Éirinn), Tim Thompson (Maynooth University), Garrett Murray (Enterprise Ireland), Niamh Kennedy (National Student Engagement Programme), Buster Whelan (Trinity College Dublin Students’ Union), Eamon Costello (Dublin City University), Ann Riordan (University College Cork). Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning HEA Generative AI Policy Framework https://hub.teachingandlearning.ie/genai/policy-framework HEA Generative AI Resource Portal https://hub.teachingandlearning.ie/genai/ Generative AI in Higher Education Teaching & Learning: Principles for Ethical AI Adoption Version 1.0, December 2025 DOI: 10.82110/qmt6-jw48 Higher Education Authority, Dublin How to cite: O’Sullivan, James, Colin Lowry, Ross Woods & Tim Conlon. Generative AI in Higher Education Teaching & Learning: Principles for Ethical AI Adoption. Higher Education Authority, 2025. DOI: 10.82110/qmt6-jw48. This document, and all original content contained within, is licensed under the Creative Commons AttributionShareAlike 4.0 International Public License (CC BY-SA 4.0). 1 1 2 6 2.1 Generative AI Adoption in Irish Higher Education Provisions of the Principles for Ethical AI Adoption Academic Integrity, Transparency, and Accountability 7 2.1.1 Institutional policies and academic freedom 7 2.1.2 Assessment design and authenticity 9 2.1.3 Disclosure, citation, and authorship 11 2.1.4 Detection and investigation 13 2.1.5 Capacity building 14 2.1.6 Governance, monitoring, and review 16 2.1.7 Summary of Recommendations 17 2.2 Equity & Inclusion 19 2.2.1 Commitment to inclusive education 19 2.2.2 Equitable access to AI tools and infrastructure 19 2.2.3 Assessment equity and standardisation 20 2.2.4 Mitigating bias and discrimination in AI systems 20 2.2.5 Irish language and minority language contexts 21 2.2.6 Supporting diverse learners and needs 22 2.2.7 AI literacy and professional development for equity 23 2.2.8 Monitoring, accountability, and redress 24 2.2.9 Summary of Recommendations 24 2.3 Critical Engagement, Human Oversight, and AI Literacy 26 2.3.1 Embedding AI literacy as core competency 26 2.3.2 Human oversight in pedagogical processes 27 2.3.3 Developing critical AI engagement 28 2.3.4 Student Development Pathways 30 Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning Table of Contents HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 2.3.5 Institutional infrastructure for AI literacy 31 2.3.6 Evaluation of AI Literacy 31 2.3.7 Governance and Accountability 32 2.3.8 Summary of Recommendations 33 2.4 Privacy & Data Governance 34 2.4.1 AI sovereignty 34 2.4.2 Data protection 35 2.4.3 Impact assessments and special categories 36 2.4.4 Approved tools and vendor governance 37 2.4.5 Transparency and intellectual property 37 2.4.6 Summary of Recommendations 38 2.5 Sustainable Pedagogy 39 2.5.1 Environmental sustainability and resource consumption 39 2.5.2 Preserving foundational intellectual capabilities 40 2.5.3 Long-term educational ecosystem health 41 2.5.4 Institutional capacity and resilience 41 2.5.5 Monitoring and continuous improvement 42 2.5.6 Summary of Recommendations 42 3References 44 Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 1 Generative AI Adoption in Irish Higher Education 1 HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 2 The Irish higher education sector has already taken important first steps toward framing the opportunities and challenges of generative AI (gen AI). The initial set of ten recommendations for gen AI adoption published by the Higher Education Authority (HEA) provided institutions with an accessible orientation for policy and practice.1 Those principles captured the immediate questions facing the sector – ranging from AI literacy and academic integrity to issues of privacy, sustainability, and sovereignty – and opened a national dialogue on how best to respond to the challenges posed to education by gen AI. The following principles, which frame the national policy framework set out by this and accompanying documents, are a direct continuation of that work: (1) Academic integrity, transparency, and accountability; (2) Equity and inclusion; (3) Critical engagement, human oversight, and AI literacy; (4) Privacy and data governance; (5) Sustainable pedagogy. These principles establish a set of core ethical commitments that should guide all institutional decisions on AI adoption, providing a common baseline for Irish higher education institutes (HEIs) that is aligned with both European regulatory requirements and wider international standards of responsible AI use. Taking such a values-led approach is essential because gen AI, particularly large language models (LLMs) like ChatGPT, radically reshapes how knowledge is produced and the validity with which it can be assessed. Decisions about whether to integrate gen AI into teaching and learning cannot be reduced to questions of efficiency or competitiveness. They must be framed in terms of educational purpose, what kind of learning experiences we want to foster, what skills and dispositions students should develop, and what forms of scholarly engagement we want to protect. Values-based guidance ensures coherence across institutions. Without shared principles, adoption risks becoming fragmented, with some HEIs embracing permissive experimentation while others default to restrictive prohibition, creating confusion for students and undermining trust. A principled stance ensures that generative AI is aligned with the mission of higher education as a public good, protecting human dignity, advancing equity of access, and safeguarding the intellectual integrity upon which democratic societies depend. These provisions are therefore not abstract aspirations but operational standards. They are designed to anchor institutional choices in fairness, transparency, accountability, and human agency, while also addressing the sectorspecific challenges of workload, assessment, and student engagement. They serve as an ethical compass for educators, ensuring that experimentation with AI remains accountable to the values that define higher education. 1O’Sullivan and Lowry, ‘Ten Considerations for Generative Artificial Intelligence Adoption in Irish Higher Education.’ Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 3 By adopting this values-led framework, HEIs will be positioned to integrate generative AI in ways that strengthen, rather than erode, the trust placed in universities. The ethical adoption of generative AI must be approached with rigour and integrity. It is not sufficient for institutions to express a commitment to ethical principles while, in practice, implementing tools or practices that compromise those very values. Sustained alignment between stated values and institutional actions is essential to ensuring that the use of generative AI strengthens, rather than diminishes, the trust placed in higher education. An institution’s approach to artificial intelligence can only be regarded as ethically robust if it demonstrates coherence between the systems it adopts and the values articulated in this framework. Where misalignment occurs, there is a risk of exposing students and staff to harm and of diminishing public confidence in higher education as a trusted steward of knowledge, integrity, and the public good. The credibility of the higher education sector rests on the consistent alignment of institutional behaviour with declared values. This entails not only compliance with relevant regulations but also active discernment in determining which tools are adopted, how they are deployed, and whether their use genuinely advances the educational mission. Where tools cannot be demonstrated to meet these ethical standards, their use should be reconsidered, and where risks are identified, they should be managed transparently and with appropriate oversight. These values are fundamental rather than discretionary, and they constitute the standards against which institutional decisions on generative AI should be measured. The provisions outlined in this framework are designed to ensure that higher education in Ireland engages with gen AI in ways that are not only innovative but also credible and trustworthy. They set the clear expectation that the integration of generative AI into teaching and learning will be conducted in a manner consistent with the highest ethical commitments of the sector. Each of the five principles addresses a distinct domain of responsibility, but they function as interdependent dimensions of a single framework. No principle operates in isolation. Decisions made under one create obligations and constraints for the others, and their collective strength derives from mutual reinforcement rather than independent application. Mapping these interdependencies shows how academic integrity relies on equity, literacy, privacy, and sustainability provisions, how equity cannot be separated from the other four domains, and how each principle shapes and is shaped by the others. Understanding these relationships is essential for coherent implementation, as institutions cannot pursue compliance in one area while neglecting its implications elsewhere. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption Beyond the specific interconnections between individual principles, certain requirements and themes recur across the entire framework. These cross-cutting elements represent those recommendations that might be considered as the core foundations of ethical AI adoption and the recurring obligations that every implementation decision should satisfy: (1) Human accountability cannot be delegated: Human judgement must be preserved in assessment, oversight, and all determinations relating to teaching and learning. (2) Legal compliance is the minimum, not the goal: Compliance with GDPR and equality legislation establishes the floor, while ethical practice determines the ceiling. (3) No AI tool is ethically complete: Openly acknowledge and document limitations and compromises and, where possible, implement compensating safeguards. (4) Transparency is a foundational requirement: Make gen AI use, procurement criteria, data practices, and system operations visible to all stakeholders. (5) Both adoption and refusal should be critical: Thoughtful rejection of gen AI tools is as valid as critical integration when pedagogical judgement demands it, but institutions should make clear to all staff that complete refusal is technically impracticable given AI’s embedding in mainstream software and workflows. (6) Support and enable staff: Institutions have a responsibility to equip staff with the knowledge and support required to engage with generative AI confidently and ethically. Professional development should be sustained, accessible across all roles, and integrated into existing frameworks for workload and recognition, regardless of whether individual staff intend to implement, investigate, or caution gen AI in teaching and learning. (7) Discipline-specific AI literacy: Staff and students across all programmes should develop a critical understanding of generative AI, encompassing its technical foundations, disciplinary applications, ethical implications, and broader societal impacts. Literacy should not be optional or confined to technical fields. (8) Students remain responsible for work they submit: AI assistance does not transfer accountability. Students answer for the accuracy and integrity of all submissions. (9) Institutional infrastructure and resourcing: Institutions should maintain adequate infrastructure and capacity to support the ethical and effective use of generative AI, including appropriate licences, governance and compliance mechanisms, and monitoring systems. (10) AI sovereignty and vendor independence: Institutions should safeguard their sovereignty over data, systems, and decisionmaking, and take active steps to avoid vendor dependencies that could compromise institutional autonomy. (11) Embed equity in every decision: Equity is not a standalone concern but a dimension of every policy choice, procurement decision, and pedagogical practice. (12) Procurement as ethical governance: Tool approval processes should verify compliance, assess bias, ensure equitable access, and evaluate environmental impact. 4 Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 11 can shift effort from learning to documentation. There are equity risks here, as students with premium features or better devices gain advantages unless institutions provide access or alternatives. Privacy and governance requirements are not embedded in the scale, yet EU and Irish guidance make data protection, transparency, and human oversight first-order duties. At higher bands students may optimise for prompt craft rather than disciplinary mastery, unless rubrics re-centre intended outcomes and keep the teacher firmly ‘in the loop’. The scale travels well into design, coding, and studio contexts but poorly into disciplines where unaided performance is essential, such as clinical judgement or initial language acquisition. Furthermore, model capabilities evolve quickly, and what counted as ‘limited support’ yesterday may now amount to substantive authorship, requiring institutions to version their implementations, schedule annual reviews, and update exemplars accordingly. For these reasons, HEIs should support a systematic renewal of assessment design to reduce substitution risks while preserving validity and equity. Programme teams should prefer tasks that make authorship and judgment visible through activities such as staged drafts with feedback, unique datasets, supervised artefact creation, oral explanations, code walkthroughs, lab or studio notebooks, or viva voce.4 Where traditional written tasks are retained, briefs should incorporate situated or reflective elements that require human reasoning tied to taught materials. Only by combining baseline institutional rules with discipline-sensitive literacy and design can higher education maintain integrity and fairness in the age of generative AI. Designing new forms of authentic or process-based assessment requires time, training, and, in some cases, additional infrastructure. Without structured institutional support, these demands risk being absorbed into current workload pressures, with potential consequences for both quality and sustainability. Policy should therefore ensure that changes to assessment design and delivery are recognised within workload models, appropriately costed in programme planning, and supported through professional development and technical assistance. AI-resilient assessment requires collective commitment and planning to sustain staff capacity and protect student learning outcomes. 2.1.3 Disclosure, citation, and authorship The integration of generative AI into academic work necessitates a fundamental reconceptualisation of authorship, attribution, and intellectual responsibility. Where traditional academic integrity frameworks assumed human-only production, the advent of gen AI creates new categories of contribution that must be made visible and assessable. HEIs should establish disclosure and citation requirements that preserve the chain of intellectual accountability while enabling legitimate AI-supported learning. Where AI is used within permitted bounds, students should include a comprehensive AI use declaration at the point of submission.5 Any such declaration is a substantive component of academic integrity that enables assessors to evaluate the authenticity and quality of student work. The declaration must be specific enough to support verification and must identify the exact name and version of the AI system used, including any plugins, 4 5 For a discussion of AI-resilient assessment practices, see supporting instruments. See established style guides on AI citation practices, which continue to develop. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 12 extensions, or additional features employed. Students must provide a clear description of how the AI was used, whether for ideation, research synthesis, code generation, language editing, or visual creation. The declaration should include representative prompts or describe the general prompting strategy employed, sufficient for assessors to understand the nature of human-AI interaction. Crucial to the declaration is a detailed account of the verification process. Students should specify the steps taken to validate gen AI outputs, including fact-checking methods and corrections made. They must also provide an honest evaluation of the relative contributions of human and AI to the final work, expressed in terms that map directly to the assessment criteria. This transparency allows assessors to calibrate their evaluation appropriately and ensures students remain conscious of their own intellectual contribution throughout the process. These declarations should be incorporated into the work itself, not relegated to separate documents that may become detached. For written assignments, the declaration should appear immediately after the title page or in a designated section of the methodology. For code submissions, it should be included in documentation headers or README files. For creative works, it should form part of the artist’s statement or design rationale. This integration ensures the declaration travels with the work through all stages of assessment and review. AI-generated content must be cited according to evolving scholarly conventions that recognise both the tool and the human operator as agents in the creative process.6 HEIs should adopt and maintain citation guidelines that distinguish between levels of assistance, from surface-level editing to substantial content generation to collaborative iteration. These distinctions matter because they signal different degrees of intellectual contribution and allow assessors to evaluate work appropriately. Where AI summarises or paraphrases existing sources, those original sources must still be cited directly, not laundered through AI attribution. This dual citation requirement ensures the intellectual genealogy of ideas remains traceable and that original authors receive proper credit. For work involving multiple rounds of humanAI interaction, citation should capture the evolutionary nature of the collaboration, documenting how ideas developed through iterative refinement. Where relevant to reproducibility or assessment, citations should include model parameters, temperature settings, or other technical specifications that shaped the output. Citation formats must evolve with disciplinary norms. While humanities disciplines may emphasise narrative descriptions of AI use that situate the technology within broader methodological frameworks, STEM fields may require more technical documentation including version control logs, prompt repositories, or computational notebooks that enable reproducibility. Professional programmes must align citation practices with sector standards, ensuring students develop habits transferable to practice. This disciplinary sensitivity prevents citation requirements from becoming either meaninglessly generic or inappropriately prescriptive. Students are the authors of work they submit and bear full responsibility for its content, accuracy, and integrity. This fundamental principle does not change when AI tools are used within permitted bounds. Students are obligated to verify all AI-generated claims against authoritative sources, with particular attention to statistical data, historical events, scientific facts, and citations that AI systems are known to fabricate. They must identify and 6 It is advised that teaching staff keep up-to-date with established academic style guides on matters relating to gen AI citation practices, which continue to develop. Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 13 address biases in gen AI outputs, including stereotypes, cultural misrepresentations, and discriminatory framings that may reflect the limitations of training data. Students must be aware, and respond appropriately, to the reality that many gen AI models have been trained on and can reproduce copyrighted material without attribution. Most critically, they must demonstrate understanding of all submitted content through the ability to explain, defend, and elaborate on any aspect of the work. This means that errors, inaccuracies, or misrepresentations in AI-generated content will be assessed as if they were the student’s own mistakes, maintaining the essential link between submission and accountability. This responsibility extends beyond fact checking and includes critical engagement with AI outputs. Students must be able to justify why they accepted, modified, or rejected AI suggestions, demonstrating the kind of intellectual autonomy that distinguishes human learning from mechanical reproduction. The capacity to evaluate AI outputs critically, to recognise their limitations, and to improve upon their outputs represents a new form of academic literacy that should be developed and assessed across disciplines. Academic staff should model the same disclosure and citation standards in all teaching materials, research outputs, and administrative documents. When lecture slides, handouts, reading lists, or sample solutions incorporate AI assistance, this must be clearly attributed. Even in routine communications such as email responses or discussion board posts, substantive use of AI assistance should be acknowledged. In research supervision, supervisors must declare any AI use in reviewing student work or generating supervisory feedback. This transparency normalises disclosure practices and maintains the mutual accountability that underpins academic relationships. Staff cannot credibly enforce standards they do not themselves observe, and their modelling of good practice provides students with concrete examples of how to integrate AI responsibly into academic work. To ensure disclosure and citation requirements are meaningful rather than performative, institutions must establish verification mechanisms that balance rigour with practicality. Quality units should periodically sample submitted work to verify that AI disclosures accurately reflect the level of assistance received. Where institutions use metadata, version histories, or interaction logs to corroborate disclosure statements, such technical verification must comply with data-protection requirements and must not serve as the sole basis for integrity determinations. Students must be aware that they may be required to explain their AI use in viva voce examinations or portfolio defences, demonstrating understanding of both the tool’s contribution and their own intellectual process. 2.1.4 Detection and investigation AI-related misconduct differs from traditional plagiarism. Whereas plagiarism involves the unacknowledged reuse of existing material, generative AI can create novel outputs that nonetheless breach traditional notions of authorship and intellectual property. Institutions must therefore re-examine investigative practices with these new conditions in mind. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 14 AI detection tools cannot provide a reliable solution. AI-use indicators and content-classification systems produce probabilistic scores rather than definitive findings, are vulnerable to evasion, and are prone to both false positives and false negatives.7 These systems typically rely on proxies like perplexity and burstiness, metrics of predictability and sentence variation, which may have been marginally useful against earlier models but fail against contemporary systems that convincingly mimic human rhythm and style. The consequence is human work written in more formal or academic registers is disproportionately flagged as AI, while AI-generated text prompted for casual tone often escapes notice. Students who write in formal or conventionalised styles, particularly non-native speakers, are disproportionately flagged,8 while sophisticated gen AI users may escape detection through prompt engineering or hybridising AI and human work. Tools trained on older models degrade quickly as new systems emerge, creating an unwinnable technological race. Beyond the statistical and technical limitations, detection tools operate on assumptions about what ‘real’ writing should look like, treating fluency, coherence, and formal register as potential signs of artificiality. By embedding narrow norms of natural writing, these tools reproduce structural inequities, rewarding those whose style falls outside the training-data uncanny valley and punishing those whose work most closely resembles the genres that models were trained on. The result is a perverse inversion of academic values, wherein clarity and control are treated as suspect, while mediocrity or idiosyncrasy passes unnoticed. This reveals why detection is not merely unreliable but actively harmful: it shifts the burden of proof onto the student, undermines trust in academic judgment, and reinscribes linguistic and cultural bias under the guise of neutrality. For these reasons, AI detection systems are not recommended as primary evidence in higher education, and serve no role as sole or determinative proof of misconduct. Investigations must rest on principles of natural justice and procedural fairness. Students must be afforded the presumption of innocence, with the burden of proof borne by the institution. Findings must be reached on the balance of probabilities, using clear, documented evidence. Responses must be proportionate to the seriousness of the allegation: suspected misuse in a minor formative task cannot be treated in the same way as substitution in a capstone or professional placement. Timeliness is also essential: cases must be progressed promptly to minimise stress and preserve learning opportunities, while allowing sufficient time for a thorough review. Evidence must be triangulated. Institutions should give primary weight to process evidence, such as draft histories, version logs, and research trails that show natural development of work. Comparative analysis across a student’s portfolio may provide useful context. Technical indicators such as metadata or citation anomalies may be considered, but only as supplementary signals. Circumstantial evidence, such as the absence of drafts, must be weighed carefully, since legitimate alternative explanations are always possible. A critical safeguard is the oral or live assessment. Where a staff member has credible grounds to believe that a submitted assessment does not represent the student’s own work, the student should be offered the opportunity to demonstrate authorship directly. Institutions should develop a policy provision that allows for any student across all modules to be called, at the request of the module coordinator, to an oral examination, 7Otterbacher, ‘Why Technical Solutions for Detecting AI-Generated Content in Research and Education Are Insufficient.’ 8Liang et al., ‘GPT Detectors Are Biased against Non-Native English Writers.’ Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 15 viva, code walk-through, studio critique, or equivalent dialogic exercise in which they account for and extend their submitted work. The policy should clearly state that, in all instances, oral assessment overrides the written artefact: if the student can demonstrate understanding, reasoning, and command of sources or methods, the oral performance confirms authenticity and secures credit. Conversely, inability to explain or extend the work constitutes strong evidence of inauthentic authorship and may warrant sanction. To ensure fairness and consistency, every such examination may involve the staff member who made the referral, but also, a panel that includes other academic staff with relevant expertise, at least one of whom must come from outside the examining unit, so that decisions are not made unilaterally and students are protected from arbitrary referral or judgement. Institutional policies of this nature do not preclude staff from having their own, module-level oral and live examination processes, but equally, the presence of such does not negate any institutional provisions. Investigations and oral examinations must be conducted fairly and transparently. Students must be informed in writing of the precise concerns, the evidence under review, their rights, and the supports available. Decisions must be communicated in writing, with reasons clearly explained and rights of appeal available. Appeals must be heard by independent staff not involved in the original decision and resolved within defined timeframes to protect student progression. All staff involved should have the opportunity to be trained in the limits of detection software, evidentiary standards, interviewing techniques that are rigorous but non-intimidating, unconscious bias awareness, GDPR compliance, and recognising when education rather than punishment is the appropriate outcome. Training should be refreshed regularly, and supported by practical case studies and shared sectoral exemplars to ensure consistency across faculties. Institutions should ensure that investigation processes are adequately supported through appropriate time allocations within workload models, access to administrative assistance, and explicit recognition within institutional planning and professional development frameworks. The purpose of investigation is not only enforcement but the preservation of trust in assessment and the protection of fairness. Sanctions have a place where deliberate deception is proven, but the primary goal remains developmental, in helping students understand expectations, protecting those who use AI responsibly and transparently, and ensuring that Irish higher education awards remain credible and trusted. 2.1.5 Capacity building The integration of gen AI into higher education depends on the capacity of staff and students to engage with these tools critically, ethically, and effectively. This capacity must not be assumed and requires systematic and sustained institutional support. HEIs should implement mandatory staff development programmes addressing both technical understanding and pedagogical practice. All teaching staff must complete core training that provides a conceptual grasp of how generative AI works, including how training data shapes outputs and embeds bias, the capabilities and limitations of current systems, and the distinctions between text, image, code, and multimodal models. Training should also cover the ethical implications of bias, opacity, and environmental impact. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 16 Assessment design represents the most urgent training need, and staff must be able to design authentic assessments that resist substitution while remaining inclusive. This includes learning to balance security and pedagogical goals, to construct rubrics that account for disclosed AI assistance, and to write clear AI-use statements for briefs. Staff should understand the workload implications of such assessments and plan accordingly, with workload models recognising the additional time required. Capacity building must also extend to integrity procedures. Staff involved in misconduct investigations should be trained in evidentiary standards, the limitations of detection tools, the triangulation of multiple evidence sources, and fair interviewing techniques. They must also understand GDPR and intellectual-property obligations, and be able to distinguish between deliberate misconduct and misunderstandings. Student induction at all levels should include structured AI literacy. Students must develop a conceptual understanding of generative AI, awareness of system limits, and skills in verifying and critiquing outputs. They must also be taught disclosure and citation practices, privacy and security responsibilities, and discipline-specific standards for ethical use. AI literacy should be embedded across curricula through integrated assignments, reflective portfolios, and case studies, not treated as a standalone module. HEIs should maintain the systems and structures necessary to support sustained capacity-building. This may include designated coordination for AI education, shared repositories of exemplars and guidance, sandbox environments for safe experimentation, and accessible support for both technical and pedagogical queries. Related content should be reviewed regularly to reflect technological developments, regulatory updates, and sectoral feedback. Adequate institutional provision is essential. Workload planning, equitable tool access, and coordinated development of expertise and systems are necessary to meet policy objectives. 2.1.6 Governance, monitoring, and review Effective governance of AI in higher education requires clear academic ownership, systematic monitoring, and responsive review mechanisms. Ultimate authority over institutional AI policy in teaching and learning should rest with the academic council or equivalent body, which should adopt institutional policies, approve major revisions, and ensure adoption supports institutional mission and strategy while keeping pedagogical decisions grounded in academic judgement. Governance bodies should receive regular reports on risk assessment covering academic, reputational, and operational dimensions, and should advise on resource priorities to ensure resources align with academic aims. Academic units are responsible for contextualised implementation, adapting institutional policy to disciplinary contexts, ensuring all assessment briefs include clear AI statements, and providing discipline-specific training and development. Units should also coordinate programme-level AI literacy to ensure progression and coherence, while leading innovation through pilot projects in AI-enhanced pedagogy. Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 17 Central support services must provide the assurance and infrastructure necessary for consistent implementation, maintaining a register of approved tools, undertaking institutional procurement and security reviews, managing licencing and equitable access, providing staff and student training, and conducting regular audits of compliance that include sampling assessment briefs for compliance and reviewing integrity cases for fairness and consistency. Institutions should introduce proportionate compliance monitoring to track implementation and support continuous improvement. Indicators might include the proportion of modules containing compliant AI statements, participation rates in staff training, and the timeliness of integrity case resolution. These should be complemented by quality measures that assess effectiveness – such as clarity of policy communication, staff confidence, and evidence of pedagogical innovation. Continuous monitoring of risk indicators should include the proportion of AI-related integrity cases, appeals, data protection incidents, and vendor compliance issues. Regular reporting contributes to transparency and accountability. Institutions are encouraged to provide an annual summary to academic governance bodies outlining compliance, quality, risk, and impact. Annual review should incorporate monitoring evidence, technological advances, regulatory updates, and stakeholder consultation, ensuring approved tools continue to meet instructional requirements with retirement procedures invoked where necessary. Continuous updates to guidance and resources should reflect current practice, with trigger-based reviews initiated in response to major capability breakthroughs, regulatory changes, significant incidents, or shifts in institutional strategy. HEIs must ensure transparency through publicly accessible AI policies, a maintained register of approved tools, and publication of summary data on use and impact. Institutions should also establish clear contact points for queries and complaints, and maintain open channels for communicating policy updates. Accountability structures should assign responsibilities unambiguously, specify decision rights, and maintain clear and transparent escalation procedures. 2.1.7 Summary of Recommendations (1) Publish a single institutional AI policy setting permitted and prohibited uses across teaching and assessment, with discipline-sensitive exemplars and protection of academic freedom. (2) Mandate institutional approval for tools so that only AI systems that pass procurement, GDPR/data-protection, and transparent, documented ethics reviews may be required for student use. (3) Maintain a public register of approved tools, updated regularly with review criteria, risks, safeguards, monitoring plans, and retirement decisions. (4) Provide institutional supports for professional development, repositories of exemplars, advisory services, equitable access to approved tools, and workload recognition. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 18 (5) Require disclosure of AI use through a standard declaration specifying tools, purpose, extent, and verification of outputs. (6) Mandate citation of AI outputs and sources, distinguishing levels of assistance and aligning with disciplinary conventions. (7) Preserve student accountability so students remain fully responsible for accuracy and integrity of submitted work, and staff must model disclosure in teaching and supervision. (8) Embed AI-use statements in every assessment brief, specifying permissions, disclosure duties, marking implications, and, where necessary, alternative non-AI pathways. (9) Adapt and further develop validated frameworks (eg. AI Assessment Scale)9 to signal allowable AI use, supplemented by institutional rules on gen AI governance. (10) Redesign assessments for authenticity, favouring approaches that make authorship and judgment visible. (11) Resource assessment reform through workload allocation and programme planning supports. (12) Ensure disciplinary sensitivity, supporting programme teams to adapt assessment rules to field-specific norms while upholding baseline integrity and equity standards. (13) Integrate AI literacy across programmes, enabling staff and students to critically evaluate AI in the context of their specific disciplines. (14) Prohibit the use of AI detectors and probabilistic tools as determinative evidence of misconduct. (15) Ground integrity investigations in natural justice, including presumption of innocence, balance of probabilities, proportionate sanctions, and timely resolution. (16) Triangulate evidence in investigations, using drafts, process records, and oral demonstrations. (17) Provide an institution-wide oral assessment safeguard wherein students must be able to demonstrate authorship live, with oral performance overriding written artefacts. (18) Ensure fairness and transparency in investigations, including written notice of concerns, access to evidence, rights of appeal, and involvement of independent panel members. (19) Train and resource investigators in evidentiary standards, unconscious bias, GDPR, interviewing, and workload recognition. (20) Establish governance, monitoring, and review, ensuring academic councils hold ultimate authority and institutions publish annual reports, conduct annual and trigger-based reviews, implement clear accountability lines, and resource governance functions adequately. 9Perkins et al., ‘The Artificial Intelligence Assessment Scale (AIAS): A Framework for Ethical Integration of Generative AI in Educational Assessment.’ Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 19 2.2 Equity & Inclusion The OECD warns that the unchecked spread of AI in education risks widening existing disparities.10 Access to AI tools risks creating new forms of educational stratification that could exacerbate inequalities within society. This principle requires that HEIs actively mitigate differential access to AI technologies and ensure that their implementation does not privilege certain learning approaches, linguistic backgrounds, or socioeconomic positions. Equity goes beyond access, requiring that AI systems be scrutinised for embedded biases that might perpetuate historical disadvantages, particularly for communities traditionally underrepresented in higher education. 2.2.1 Commitment to inclusive education The adoption of AI in higher education must be aligned with Ireland’s obligations under equality legislation, the Public Sector Equality and Human Rights Duty,11 and the national commitment to inclusive education under the UN Sustainable Development Goal 4, to ‘ensure inclusive and equitable quality education and promote lifelong learning opportunities for all’. International guidance reinforces this, such as UNESCO’s global frameworks on AI in education which stress that AI adoption must be grounded in humanistic values, including inclusion, equity, gender equality, and respect for cultural and linguistic diversity.12 Every HEI should include within its institutional AI policy a clear statement affirming equity and inclusion as guiding principles. This commitment should be accompanied by a transparent account of how it will be implemented through procurement processes, staff and student development, and curriculum design, ensuring that no cohort is disadvantaged in an AI-enabled learning environment. The commitment should also acknowledge intersectional disadvantage, recognising that students experiencing multiple forms of marginalisation face compounded rather than additive barriers. 2.2.2 Equitable access to AI tools and infrastructure Equitable access to AI is not guaranteed. Many gen AI tools require reliable broadband and paid subscriptions. Students from lower socio-economic backgrounds or from regions with limited connectivity are at risk of being excluded from AI-enabled learning, so institutions should take proactive steps to ensure that engagement with AI does not depend on students’ private means. This includes providing institutional licences for approved tools and ensuring that campuses are equipped with the hardware, software, and connectivity needed to support inclusive access, alongside mechanisms to provide individual access where necessary. The risk of digital poverty extends beyond hardware and subscriptions. Students experiencing housing instability may lack quiet spaces for AI-assisted study. Working students may have limited time to develop AI literacy skills. Commuter students may struggle with campus-based AI resources. Institutions must recognise 10 Varsik and Vosberg, ‘The Potential Impact of Artificial Intelligence on Equity and Inclusion in Education.’ 11 ‘Public Sector Equality and Human Rights Duty.’ 12 Miao and Holmes, ‘Guidance for Generative AI in Education and Research.’ HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 20 these broader dimensions of digital exclusion in support strategies that address the full spectrum of access barriers. Possible measures include extending library access with dedicated AI-enabled study spaces, offering asynchronous AI-literacy training that accommodates work schedules, and ensuring that AI tools are mobileoptimised for students who rely primarily on smartphones. 2.2.3 Assessment equity and standardisation The use of AI in assessment contexts presents particular equity challenges. Students with greater financial means might access more sophisticated AI tools for take-home assignments, creating unfair advantages over peers relying on free or institutional versions. Institutions should be alert to this emerging digital divide in assessment and take steps to preserve fairness and integrity. Institutions should establish clear protocols for the use of AI in assessment that take account of differential access among students. Where AI use is permitted, HEIs may either designate specific, institutionally provided tools or require students to submit detailed declarations outlining which AI tools were used and for what purposes. Assessment rubrics should be adapted to evaluate critical engagement with AI rather than simply the sophistication of AI-generated content. For time-bound assessments, institutions might consider providing standardised AI access through controlled environments, ensuring all students work with the same tools under the same conditions. Alternative assessment strategies that minimise advantage from differential AI access should be prioritised. These might include in-person presentations, reflective portfolios that document learning processes, collaborative projects where AI use is transparent and shared, or hybrid assessments combining AI-assisted preparation with non-AI demonstration of understanding. 2.2.4 Mitigating bias and discrimination in AI systems AI systems are not neutral. They are trained on large datasets that frequently contain social biases related to race, gender, class, language, and disability. These biases, if left unaddressed, can be reproduced or amplified in outputs. The OECD has highlighted that unchecked adoption of AI can entrench inequities and undermine cultural responsiveness,13 while the European Commission’s guidelines for trustworthy AI explicitly identify diversity, non-discrimination, and fairness as key requirements for responsible practice.14 HEIs have a duty to ensure that the AI systems they approve for teaching and learning are subject to rigorous scrutiny for bias and discrimination. It is not sufficient for institutions to rely on vendor assurances or generic claims of compliance. Instead, this obligation must be operationalised through procurement and approval processes that apply explicit equity criteria to every AI system under consideration. These processes should require clear evidence of transparency in the provenance of training data, enabling institutions to assess 13 Varsik and Vosberg, ‘The Potential Impact of Artificial Intelligence on Equity and Inclusion in Education.’ 14 ‘Ethics Guidelines for Trustworthy AI.’ Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 27 They should also cultivate ethical reasoning concerning authorship, accountability, equity, and the preservation of human judgement, as well as practical capabilities in prompt design, verification, and collaborative human–AI workflows. Generic digital-skills provision alone is insufficient. A medical student must understand how AI diagnostic tools interact with clinical judgement and patient care ethics, an engineering student requires knowledge of how AIgenerated designs must be validated against safety standards and professional liability, and a humanities student must engage with how AI text generation relates to authorial voice, interpretative traditions, and cultural production. For this reason, AI literacy should therefore be scaffolded across the stages of a programme, progressing from introductory awareness to advanced critical application, with definitions of ‘advanced’ determined by teaching staff within their disciplinary context. This progression depends on robust, disciplinespecific AI literacy training for staff, ensuring that educators have the knowledge and confidence to define and assess critical engagement in their fields. At introductory levels, students might engage the technology through guided reflection and low-stakes experimentation, comparing their own reasoning with AI outputs and identifying limitations or errors. At intermediate stages, analysis could extend to comparing responses across multiple systems, experimenting with prompt design, and exploring how training data influences interpretation in their discipline. By later stages, students should demonstrate sophisticated critique and application, whether through research on AI’s professional implications, applied use cases, or evaluation of the epistemological consequences of machinegenerated outputs. Programme teams should ensure that this progression is coherent and deliberate, with learning outcomes and assessment strategies mapped to promote cumulative development. Students should also encounter AI through interdisciplinary perspectives that highlight its wider societal significance. Cross-programme seminars, collaborative projects, and integrated modules should bring together perspectives from technology, humanities, social sciences, and professional studies. Joint teaching initiatives and guest lectures from diverse practitioners and researchers should, where feasible, be embedded within formal curricula rather than confined to optional enrichment. Embedding both disciplinary depth and interdisciplinary breadth ensures that graduates emerge not only with functional skills but with the conceptual, critical, and ethical understanding required to navigate an AI-saturated world. 2.3.2 Human oversight in pedagogical processes The EU AI Act’s designation of educational AI as high-risk systems mandates human oversight, recognising that educational decisions shape human potential in ways that require moral accountability. Article 1423 requires oversight mechanisms that minimise risks to fundamental rights, ensure systems are used as intended, and allow human intervention or discontinuation when necessary. This means academic staff should retain ultimate responsibility for pedagogical decisions. This principle is most critical in assessment, where grading requires professional judgement that algorithms cannot replicate. Decisions about whether work meets the required standard, or how effectively a student has demonstrated 23 https://artificialintelligenceact.eu/article/14/ HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 28 understanding, must rest with academic staff who can consider disciplinary context and individual circumstances. Institutions should maintain clear protocols to ensure that all evaluative decisions about student work are made by humans, preserving the essential role of academic expertise in recognising and validating learning. Comparable oversight is needed in curriculum design. Adaptive learning platforms that promise personalisation may inadvertently narrow intellectual horizons. When algorithms optimise for engagement or completion rates, they may steer students away from difficult concepts or controversial topics that are essential to disciplinary understanding. Academic staff should therefore review algorithmic recommendations against programme learning outcomes to ensure that systems do not limit exposure to the full breadth of disciplinary knowledge. Effective oversight should be embedded within existing academic-governance structures. Institutions may designate committees or roles responsible for monitoring the use of generative-AI systems, ensuring compliance with oversight principles, and recommending suspension where systems fail to meet required standards. Disciplinary units could identify coordinators with understanding of both the technological and pedagogical dimensions of AI, enabling them to advise colleagues, audit local practice, and raise concerns through established reporting routes. Oversight structures should have clearly defined responsibilities and the authority to act on identified risks, including requesting documentation, commissioning bias testing, and recommending changes to practice. Locating such authority within recognised governance frameworks ensures that oversight is substantive rather than symbolic, enabling institutions to benefit from AI’s analytical capacities while safeguarding the human expertise and ethical judgement that define higher education. 2.3.3 Developing critical AI engagement Developing critical engagement with AI requires intellectual frameworks that go beyond operational competency. Students must not only learn how to use AI tools effectively but also acquire the ability to interrogate their implications and limitations. This literacy draws on philosophy, sociology, linguistics, and postcolonial critique, interrogating intelligence and consciousness, analysing power and truth, and exposing whose knowledge is privileged. Institutions must ensure students gain a conceptual vocabulary that frames gen AI as a sociotechnical system rather than a neutral tool, recognising how datasets encode historical inequities, how model architectures reflect design choices, how deployment contexts shape interpretation, and how feedback loops can amplify bias. Students should understand gen AI systems as cultural artefacts, marked by the values and blind spots of their creators and training data. Such engagement requires historical perspective. AI must be situated within the longer history of automation and social change. Students should explore how earlier technologies promised liberation but sometimes introduced new forms of control; how efficiency gains have often accrued unevenly; and how technological Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 29 determinism can obscure political choice. This perspective supports critical reflection while leaving room for responsible innovation. Disciplinary perspectives enrich this analysis: philosophy addresses agency and moral responsibility, sociology and anthropology reveal AI’s embeddedness in social relations, literary and cultural studies analyse its impact on creativity and representation, economics and political science show how AI reshapes labour markets and democracy. Together, these lenses foster comprehensive understanding of gen AI’s implications. Critical engagement also depends on robust evaluative practices. Students must move beyond binary acceptance or rejection of AI outputs and instead apply systematic protocols suited to their fields. Empirical disciplines may emphasise verification of sources and identification of data artefacts, while interpretive disciplines can focus on nuance and rhetorical coherence. All students should be equipped to detect bias, analyse how prompting influences outputs, and understand that bias operates not only technically but structurally, requiring broader societal awareness. For this vision to be realised, teaching staff need structured opportunities for professional development. Institutions should provide accessible development that builds shared foundations across disciplines, covering technical fundamentals, pedagogical implications, ethical and policy frameworks, and approaches to course integration. Progression pathways should allow staff to develop expertise in areas such as assessment design, AI ethics, technical application, or educational research, recognised through institutional or sectoral accreditation. Peer-learning networks and communities of practice can sustain development across disciplines and institutions, building collective expertise. Professional development should also address resistance and anxiety. Many educators experience uncertainty about AI or perceive it as incompatible with their professional values. Development programmes should provide space for dialogue, acknowledge concerns about workload and identity, and position AI as a tool that supports, rather than replaces, academic judgement. Low-stakes experimentation and recognition of diverse approaches can build confidence incrementally. Values-based framing should link gen AI integration directly to educational purposes, such as enhancing learning, preserving rigour, and freeing staff to focus on high-value interactions. Effective adaptation requires time and institutional support. Building AI literacy prior to curriculum or assessment redesign should be recognised within workload and development frameworks rather than treated as an additional burden. Institutions should integrate this activity into existing professional-development systems, ensure equitable access for part-time and sessional staff, and provide access to pedagogical and technical guidance. Incentives for innovation and recognition of exemplary practice can further embed a culture of critical and ethical engagement with AI across the sector. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 30 2.3.4 Student Development Pathways Developing AI literacy among students requires coherent and progressive curriculum design. Institutions should establish clear pathways that ensure all students achieve threshold competencies while creating opportunities for advanced development among those pursuing AI-intensive disciplines or careers. Foundational learning should be provided early in programmes, introducing conceptual understanding of generative AI, critical evaluation frameworks, ethical considerations, institutional policies, and practical skills for appropriate use. Progressive integration across programmes should embed AI literacy into disciplinary contexts. This progression should be mapped at programme level to ensure coherence, coverage, and the avoidance of redundancy. Cocurricular opportunities can extend learning for those who wish to specialise, through AI literacy certificates, hackathons, research assistantships, and peer tutoring. These should be accessible through varied formats, times, and entry points to promote inclusion. Pathways should accommodate the diversity of students’ backgrounds and capabilities. Students enter higher education with different levels of confidence in generative AI, from those already programming models to those with limited digital experience. Differentiated entry points can recognise prior learning while guaranteeing critical engagement for all. Importantly, coding ability cannot substitute for critical thinking, and all students should demonstrate the ability to evaluate and contextualise AI outputs. Accessibility must be built into all provision. Students with learning differences may require additional support with abstract concepts or alternative assessment approaches, while international students may require language support for technical vocabulary and cultural contextualisation of AI examples. AI literacy also entails ethical development and digital citizenship. Students should be prepared as responsible participants shaping gen AI’s societal impact. Academic integrity education should help students to understand why intellectual effort matters and what responsibilities accompany the use of generative AI. Case studies, discussions of ambiguous scenarios, and reflection on personal values can develop nuanced understanding beyond rule compliance. Students should also build data consciousness, recognising how their interactions contribute to training, what rights they have over their data, and how to evaluate privacy practices. They should be encouraged to consider social responsibility, how gen AI adoption may reshape employment in their field, what professional obligations exist for transparency, and how they can contribute to beneficial and ethical AI development. Graduates should emerge not only able to use AI but equipped to shape its direction in ways consistent with societal needs and the values of higher education. Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 31 2.3.5 Institutional infrastructure for AI literacy Effective AI literacy development depends on institutional structures that extend beyond fragmented departmental initiatives. Central coordination helps to ensure consistency, coherence, and access to expertise that individual units may not sustain alone. Institutions should consider designating a coordinating unit or network as a focal point for staff and student development in AI literacy, supported by educational developers and technical specialists. Such provision should offer curriculum resources, coordinate professional development, advise on integration, and evaluate effectiveness. Accessibility should be ensured through a combination of physical and virtual spaces that enable collaboration, workshops, and responsive support. Resource development should be systematic and sustainable. Discipline-specific exemplars, assessment templates, and interactive tutorials should be developed in partnership with academic departments, regularly updated to reflect technological change, and maintained through version control and periodic review. Quality assurance should be integrated into existing institutional frameworks, including evaluation of learning outcomes, feedback from staff and students, benchmarking against sector practices, and continuous enhancement. A reliable technological environment is essential for authentic learning and safe experimentation. Institutions should provide licensed access to approved gen AI tools, sandbox environments for exploration, and controlled API access for advanced users. These platforms should operate under clear usage policies, data protection and security protocols, and responsive support. Partnership and collaboration are central to sustainable provision. Industry links can offer students authentic experience through guest lectures, internships, and project-based collaboration, while institutions retain academic independence and critical distance. Inter-institutional cooperation allows sharing of resources, joint staff development, and collaborative research on effective practice, with sector-level coordination enhancing collective impact. Engagement with the wider community through public lectures, school outreach, and lifelonglearning initiatives extends the benefits of AI literacy beyond higher education and reinforces the sector’s civic role in shaping responsible technological futures. 2.3.6 Evaluation of AI Literacy Assessment of AI literacy should evaluate technical proficiency alongside critical thinking and ethical reasoning. Traditional testing of knowledge recall or procedural skills cannot capture the judgement that genuine literacy demands. A range of assessment approaches may be appropriate. Portfolios can allow students to demonstrate development over time through collected artifacts, reflective commentary, and evaluation of both their own and others’ gen AI use. Case-based assessments present complex scenarios where students must weigh benefits and risks of gen AI deployment or resolve ethical dilemmas. These tasks reflect authentic professional contexts rather than abstract exercises. Collaborative projects add a social dimension, requiring students to work in groups to investigate AI’s implications or use gen AI tools responsibly, with individual contributions documented through process notes and peer evaluation. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 32 Evaluation of AI literacy should also occur at programme level. Institutions should consider whether graduates achieve threshold competencies through mechanisms such as capstone projects, cumulative portfolios, or external validation by employers and professional bodies. Results from such evaluations should inform ongoing programme review, highlighting gaps, revising outcomes, and guiding resource allocation. In certain disciplines, employer feedback is particularly important, and should be gathered systematically through focus groups and analysis of placement data to ensure graduates meet professional expectations. Longitudinal studies of student progression can reveal how literacy develops over time, identifying effective practices and areas for enhancement. Given the rapid evolution of AI technologies, AI literacy provision should operate within continuous-improvement frameworks. Regular review, piloting, and the transparent dissemination of all findings guard against stagnation and ensure that AI literacy education remains responsive to technological and societal change. 2.3.7 Governance and Accountability Effective implementation of critical engagement, human oversight, and AI literacy requires both visible leadership and distributed responsibility. Ultimate accountability rests with senior leadership, while operational responsibility is embedded across all levels of the institution. Executive sponsorship signals priority and provides external advocacy. The senior sponsor should chair institutional AI governance, reporting to academic council to ensure alignment with the educational mission. Academic leadership through heads of school or unit ensures disciplinary integration, with responsibility for adapting policy to local contexts, supporting staff development, and monitoring course transformation. These leaders act as bridges between institutional strategy and departmental practice, translating policy into pedagogy and ensuring coherence between frameworks and classroom implementation. Operational coordination depends on clearly defined roles. Institutions may designate an AI-education lead to oversee literacy development and service coordination, supported by local coordinators who provide discipline-specific advice and escalate emerging issues. These roles should be recognised through workload allocation, administrative support, and inclusion within performance and development frameworks. Institutions should establish proportionate metrics that capture compliance, effectiveness, and risk. Implementation indicators might track the proportion of programmes with integrated AI literacy, the share of staff completing professional development, and the number of students achieving threshold competencies. Quality indicators should measure student confidence and staff satisfaction. Risk indicators should identify potential problems such as integrity breaches, student complaints, or technical failures. Regular reporting, both internal and public, provides transparency through annual summaries of progress and case studies of innovative practice. Independent review through external experts, benchmarking, and sector-wide participation validates standards and identifies areas for improvement. Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 33 Sustainability requires strategic planning and reliable provision of core supports. Institutions should plan for long-term investment in infrastructure, staffing, licensed tool access, and professional development. Efficiency can be achieved through shared services, scalable delivery of foundational content, and peer learning networks, provided that quality and equity are maintained. Enduring impact will depend on combining clear leadership, distributed responsibility, continuous evaluation, and sustainable planning within a coherent institutional framework. 2.3.8 Summary of Recommendations (1) Recognise AI literacy as a core graduate attribute in all programmes. (2) Define programme-specific learning outcomes covering technical foundations, disciplinary applications, critical evaluation, ethics, and practical workflow skills. (3) Map a scaffolded progression from introductory awareness to advanced critical application across stages of study, with assessment aligned to outcomes. (4) Provide discipline-specific professional development opportunities so educators can teach and assess AI literacy with confidence and credibility. (5) Embed interdisciplinary perspectives through seminars or modules linking technical, social, ethical, and cultural perspectives. (6) Codify human oversight so that academic staff retain final authority over assessment, grading, feedback, and curriculum decisions. Institutions should also ensure that students achieve programme-level AI-literacy outcomes even where specific modules restrict AI use. (7) Implement oversight protocols ensuring that all machine outputs require human review before action. (8) Monitor adaptive and personalised systems to prevent curricular ‘filter bubbles’, comparing algorithmic recommendations with programme outcomes. (9) Establish or designate an institutional oversight committee empowered to request documentation, commission bias testing, pause deployment, and discontinue non-compliant tools. (10) Identify unit-level AI-education coordinators to advise colleagues, monitor local use, and escalate concerns. (11) Include oversight requirements in procurement, covering human-in-the-loop capability, intervention and override functions, logging, declarations of fitness for purpose, and known limitations. (12) Develop student development pathways that provide foundational induction for all learners and optional advanced tracks, projects, and credentials for specialisation. (13) Differentiate entry points and ensure accessibility for varied prior experience, disabilities, and language needs. (14) Provide central coordination for AI-literacy resources, exemplars, and responsive support for staff and students. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 34 (15) Maintain secure technical infrastructure: licensed tools, sandbox environments, LMS integration, APIs where appropriate, supported by clear policies and support. (16) Acknowledge staff time for training, assessment redesign, and course transformation within workload and development frameworks, ensuring equitable provision for part-time and sessional staff. (17) Assess AI literacy authentically using portfolios, case-based tasks, and collaborative projects aligned with disciplinary contexts. (18) Evaluate outcomes at programme level using cumulative evidence and, where appropriate, input from employers or professional bodies. (19) Embed continuous-improvement processes with regular review of content, pilot initiatives, evaluation findings, and transparent dissemination of results (20) Measure and report progress on implementation, quality, and risk through annual internal and public reporting that includes student perspectives. 2.4 Privacy & Data Governance Privacy and data governance form a central pillar of responsible AI adoption in higher education, recognising the particular vulnerabilities that arise when educational data interacts with gen AI systems. Learning generates sensitive data, and any perceived efficiency gains from gen AI must not come at the expense of student privacy, autonomy, or trust. Ireland’s strong data-protection tradition establishes a high baseline of expectation. HEIs should act as exemplary custodians of the digital traces entrusted to them, embedding privacy-by-design across procurement, governance, and pedagogy. Security and transparency should extend across the full lifecycle of AI adoption, from initial approval of tools to their everyday use by staff and students. 2.4.1 AI sovereignty Institutions should ensure that their use of artificial intelligence preserves institutional sovereignty: the ability to retain meaningful control over their own data, systems, and decision-making. AI sovereignty goes beyond regulatory compliance. It is the condition that allows HEIs to act in accordance with their missions, unconstrained by the commercial or technical architectures of external vendors. AI sovereignty requires that institutional data, intellectual property, and records remain under the stewardship of the institution at all times. Vendor contracts should make explicit that data provided to or generated by AI systems belongs to the institution, that it will not be reused for training without explicit agreement, and that it can be retrieved in standard formats on termination of service. Institutions should have clear exit strategies for all critical systems, so that they can discontinue the use of a tool without losing access to, or control over, institutional records or student data. Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 35 Procurement processes should consider sovereignty alongside functionality, cost and accessibility. Preference should be given to systems that rely on open standards, that provide full documentation, and that support interoperability and portability of data. Where reliance on a proprietary model is unavoidable, institutions should document the risks of lock-in and ensure that mitigations are in place, such as time-limited contracts and contingency planning. Sovereignty also entails the ability to audit and explain the behaviour of systems used in the name of the institution. AI systems must be subject to institutional oversight, and institutions should retain the right to suspend or discontinue a system where its outputs cannot be justified to students, staff, regulators, or the public. This principle is inseparable from academic freedom: teaching, research, and administration must not be dictated by opaque gen AI models or commercial interests that override institutional judgement. The global concentration of generative-AI development within a small number of large providers presents strategic risks for higher education. Reliance on external proprietary infrastructure may limit institutional and sectoral autonomy in areas such as curriculum design, research direction, and graduate capability development. It can also expose institutions to potential changes in access, pricing, or terms of service that conflict with academic priorities or national policy objectives. Managing these dependencies requires proactive assessment of vendor relationships, transparent contractual arrangements, and the capacity to transition to alternative systems where necessary to protect academic independence and continuity of service. Generative-AI tools inevitably reflect the priorities, assumptions, and biases of their developers, shaped by the datasets, design decisions, and commercial objectives that underpin them. When adopted at scale without appropriate oversight or contextual adaptation, such systems risk narrowing curricular diversity and privileging particular perspectives. Preserving educational integrity and cultural relevance therefore requires that institutions retain the capacity to adapt, contextualise, or where necessary decline the use of AI systems that do not align with their educational missions or the values of their communities. 2.4.2 Data protection All HEIs must adopt institution-wide policies that guarantee compliance with GDPR and the EU AI Act, or other applicable data protection and AI-related legal frameworks where they apply. Legal requirements establish the minimum threshold for lawful operation, and institutions are responsible for ensuring their practices meet all relevant statutory obligations. Beyond compliance, institutions should implement best practices that reflect the sensitivity of educational data. Personal data collection and processing in the context of generative AI should be strictly necessary for clearly defined educational purposes, supported by documented justification for each category of data collected. The lawful basis for processing should be explicit, proportionate, and subject to regular review. The principle of data minimisation should guide all decisions regarding information gathering and retention. Secure storage and handling of educational data require safeguards commensurate with the sensitivity of the information held. Measures such as encryption, role-based access controls, multi-factor authentication, and comprehensive logging and monitoring represent examples of appropriate controls that can help ensure HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 36 auditability and resilience. Retention periods should align with statutory requirements and be enforced through automated technical processes wherever possible. Regular security testing, vulnerability assessments, and supplier reviews can provide assurance that controls remain effective against evolving threats, with clear procedures for addressing any findings. Transparency obligations should extend beyond legal notification to meaningful communication. Staff and students should receive plain-language explanations about data collection purposes, legal bases, categories of data, recipients, international transfers, and foreseeable consequences, both at the point of collection and throughout the lifecycle of their data. Explanations of automated processing should avoid legal jargon while remaining accurate and comprehensive, and should be reviewed periodically to ensure continued accuracy. Data-subject rights must be practically exercisable and supported through clear institutional processes. Institutions should maintain accessible procedures for individuals to request access to, correction of, restriction of, or deletion of their personal data, and for exercising the right to data portability where applicable. Response times should reflect the significance and urgency of the request. Where technical or legal constraints limit full compliance, institutions should offer proportionate alternatives and ensure that the right to object to specific processing purposes is respected without disadvantage to the individual’s educational experience. 2.4.3 Impact assessments and special categories Data Protection Impact Assessments (DPIAs) should be undertaken before the deployment or significant modification of any AI system that processes personal data in educational contexts. Assessments should explicitly consider algorithmic fairness, explainability requirements, data quality, and cumulative privacy risks arising from the interaction of multiple systems or datasets. Reviews of existing DPIAs should capture changes in processing operations, emerging risks, and evolving regulatory guidance. Where residual risks remain high after mitigation, consultation with the relevant supervisory authority should be undertaken. Special categories of personal data require safeguards reflecting their heightened sensitivity. Health data, disability accommodations, religious or philosophical beliefs, and ethnic origin data that arise in educational contexts demand explicit consent or another appropriate lawful basis with documented necessity and proportionality. Processing should be restricted to authorised personnel with appropriate training, and technical and organisational measures should prevent unauthorised access or disclosure. Regular audits should verify that enhanced protections are consistently implemented. Cross-border data transfers add further complexity and should be managed with particular care. Where gen AI systems involve international data flows, institutions should ensure that appropriate safeguards are in place. Transfers to jurisdictions without adequate data-protection standards should be subject to specific risk mitigations, and data localisation should be considered, where feasible, for sensitive educational records and special-category data. Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 43 (14) Evaluate long-term financial implications of AI adoption, including licensing, training, and infrastructure. (15) Mitigate vendor lock-in risks, favouring open standards, data portability, and clear exit strategies in procurement. (16) Incorporate resilience planning into AI strategy to ensure continuity of learning if specific tools fail or become unavailable. (17) Periodically review dependencies to identify and mitigate single points of failure in AI provision. (18) Adopt pilot-first approaches, testing innovations at small scale and using evidence-based evaluation before wider implementation. (19) Monitor educational outcomes longitudinally to assess retention, transfer, and long-term student development in AIenabled contexts. (20) Review AI adoption regularly for environmental, educational, and financial sustainability, adjusting practice based on evidence and sharing lessons across the sector. HEA | Generative AI in Higher Education in Teaching & Learning Principles for Ethical AI Adoption 44 References 3 Principles for Ethical AI Adoption HEA | Generative AI in Higher Education in Teaching & Learning 45 Digital Competence Framework for Citizens (DigComp). https://joint-research-centre.ec.europa.eu/projectsand-activities/education-and-training/digital-transfo rmation-education/digital-competence-frameworkcitizens-digcomp_en. Elhussein, Gene, Elselot Hasselaar, Ostap Lutsyshyn, Tanya Milberg, and Saadia Zahidi. ‘Shaping the Future of Learning: The Role of AI in Education 4.0.’ World Economic Forum, 2024. https://www.weforum.org/publications/shaping-thefuture-of-learning-the-role-of-ai-in-education-4-0/. Empowering Learners for the Age of AI: An AI Literacy Framework for Primary and Secondary Education (Review Draft). OECD, 2025. https://ailiteracyframework.org. Ethics Guidelines for Trustworthy AI. European Commission: Directorate-General for Communications Networks, Content and Technology and High-level Expert Group on Artificial Intelligence, 2019. https://doi.org/10.2759/346720. Kestin, Greg, Kelly Miller, Anna Klales, Timothy Milbourne, and Gregorio Ponti. ‘AI Tutoring Outperforms In-Class Active Learning: An RCT Introducing a Novel Research-Based Design in an Authentic Educational Setting.’ Scientific Reports 15 (2025). https://doi.org/10.1038/s41598-02597652-6. Liang, Weixin, Mert Yuksekgonul, Yining Mao, Eric Wu, and James Zou. ‘GPT Detectors Are Biased against Non-Native English Writers.’ Patterns 4, no. 7 (2023). https://doi.org/10.1016/j.patter.2023.100779. Miao, Fengchun, and Wayne Holmes. ‘Guidance for Generative AI in Education and Research.’ UNESCO, 2023. https://doi.org/10.54675/EWZM9535. Miao, Fengchun, and Cukurova Mutlu. ‘AI Competency Framework for Teachers.’ UNESCO, 2024. https://doi.org/10.54675/ZJTE2084. O’Sullivan, James, and Colin Lowry. ‘Ten Considerations for Generative Artificial Intelligence Adoption in Irish Higher Education.’ Higher Education Authority, 2025. https://hub.teachingandlearning.ie/genai/tenconsiderations-for-generative-artificial-intelligenceadoption-in-irish-higher-education/. Otterbacher, Jahna. ‘Why Technical Solutions for Detecting AI-Generated Content in Research and Education Are Insufficient.’ Patterns 4, no. 7 (2023). https://doi.org/10.1016/j.patter.2023.100796. Perkins, Mike, Leon Furze, Jasper Roe, and Jason MacVaugh. ‘The Artificial Intelligence Assessment Scale (AIAS): A Framework for Ethical Integration of Generative AI in Educational Assessment.’ Journal of University Teaching and Learning Practice 21, no. 06 (2024). https://doi.org/10.53761/q3azde36. Public Sector Equality and Human Rights Duty. Department of Public Expenditure, Infrastructure, Public Service Reform and Digitalisation, 2021. https://gov.ie/en/department-of-publicexpenditure-infrastructure-public-service-reform-a nd-digitalisation/organisation-information/publicsector-equality-and-human-rights-duty/. Varsik, Samo, and Lydia Vosberg. ‘The Potential Impact of Artificial Intelligence on Equity and Inclusion in Education.’ OECD, 2024. https://doi.org/10.1787/15df715b-en. Voultsiou, Evdokia, and Lefteris Moussiades. ‘A Systematic Review of AI, VR, and LLM Applications in Special Education: Opportunities, Challenges, and Future Directions.’ Education and Information Technologies 30, no. 13 (2025): 19141–81. https://doi.org/10.1007/s10639-025-13550-4. Warschauer, Mark, Waverly Tseng, Soobin Yim, et al. ‘The Affordances and Contradictions of AIGenerated Text for Writers of English as a Second or Foreign Language.’ Journal of Second Language Writing 62 (2023). https://doi.org/10.1016/j.jslw.2023.101071. Higher Education Authority 3 Shelbourne Buildings Crampton Avenue Shelbourne Road Ballsbridge D04 C2Y6 Contact Phone: +353 1 231 7100 Lo-Call Number: 1890 200 637 Fax: +353 1 231 7172 Email: [email protected] www.hea.ie @hea_irl